Research On Detection And Identification Method Of Pointer Instrument In Substation Environment

In recent years,with the increase of investment in smart grid and the continuous maturity of artificial intelligence technology,substation,as the hub of power grid,is gradually developing in the direction of intelligence.A large number of inspection robots are used in substations to ensure their safe operation.The internal environment of the substation is complex,and the images of the instrument collected by the inspection robots are easy to be disturbed,resulting in poor accuracy of automatic identification of instrument readings,which is the biggest problem faced by inspection robots in the process of automatic inspection of instruments.Therefore,this paper comprehensively studies and analyzes the position detection of the instrument,the uneven illumination correction of the instrument image as well as the indication recognition of the instrument taking the pointer instrument in the substation environment as the research object.Finally,an intelligent meter reading method with high speed and high precision is realized.The main research work of this paper is as follows:(1)A position detection algorithm for substation pointer instrument based on improved CenterNet is proposed.The algorithm selects ResNet-50 as the feature extraction network,replaces the ordinary convolution in ResNet-50 with deep separable convolution,and greatly compresses the operation parameters of the model.The instance standardization is combined with the batch standardization to form the instance batch standardization,which is applied to the bottleneck module of ResNet-50 to enhance the learning efficiency of the network for representing style information and deep content features,so as to improve the detection speed and accuracy of the algorithm for pointer instruments.(2)The pointer instrument image with noise interference and uneven illumination is preprocessed.Firstly,Gaussian filter is used to smooth and denoise the instrument image,and then the improvement of uneven illumination of instrument image is studied.By comparing several common uneven illumination correction algorithms,the correction algorithm based on Gamma-2D function is finally determined to reduce the impact of uneven illumination on the instrument image.(3)This paper designs an indicator recognition algorithm for pointer instrument.The prior information of different types of instruments to be tested shall be collected in advance to reduce the difficulty of identification.In order to reduce the interference caused by the background outside the dial area to the pointer detection,the Hough circle detection method is used to extract the instrument dial area.In the extracted dial area,the pointer detection is completed by the combination of Zhang thinning algorithm and cumulative probability Hough transform,and the angle method is selected to obtain the final instrument indication.Finally,the effectiveness of the algorithm is verified by experiments.Figure[49]table[5]reference[86]...